The Implicit Bias of Logit Regularization

• URL: http://arxiv.org/abs/2602.12039v2
• Date: Fri, 13 Feb 2026 14:03:40 GMT
• Title: The Implicit Bias of Logit Regularization
• Abstract summary: We analyze a class of logit regularizers in the context of linear classification.<n>We demonstrate that they induce an implicit bias of logit clustering around finite per-sample targets.<n>Our results extend the theoretical understanding of label smoothing and highlight the efficacy of a broader class of logit-regularization methods.
• Score: 4.588028371034407
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Logit regularization, the addition of a convex penalty directly in logit space, is widely used in modern classifiers, with label smoothing as a prominent example. While such methods often improve calibration and generalization, their mechanism remains under-explored. In this work, we analyze a general class of such logit regularizers in the context of linear classification, and demonstrate that they induce an implicit bias of logit clustering around finite per-sample targets. For Gaussian data, or whenever logits are sufficiently clustered, we prove that logit clustering drives the weight vector to align exactly with Fisher's Linear Discriminant. To demonstrate the consequences, we study a simple signal-plus-noise model in which this transition has dramatic effects: Logit regularization halves the critical sample complexity and induces grokking in the small-noise limit, while making generalization robust to noise. Our results extend the theoretical understanding of label smoothing and highlight the efficacy of a broader class of logit-regularization methods.

Related papers

• Semi-Supervised Learning with Noisy Proxy Covariates: Generalization Bounds and Distribution Regression [6.759672187503063]
  We show that fast labeled sample rates are recovered when proxy is controlled.<n>We also show that distribution regression is a direct special case, with analogous guarantees when the finite bag size is large enough.
  arXiv  Detail & Related papers  (2026-05-30T04:01:14Z)
• Provably Reliable Classifier Guidance via Cross-Entropy Control [4.298880233819988]
  We show that cross-entropy loss at each diffusion model step is sufficient to control the corresponding guidance error.<n>Our result yields an upper bound on the sampling error vectors-guided diffusion models and bears resemblance to a reverse log-Sobolev-type inequality.
  arXiv  Detail & Related papers  (2026-01-29T02:59:04Z)
• Benign Overfitting and the Geometry of the Ridge Regression Solution in Binary Classification [75.01389991485098]
  We show that ridge regression has qualitatively different behavior depending on the scale of the cluster mean vector.<n>In regimes where the scale is very large, the conditions that allow for benign overfitting turn out to be the same as those for the regression task.
  arXiv  Detail & Related papers  (2025-03-11T01:45:42Z)
• Learning a Gaussian Mixture for Sparsity Regularization in Inverse Problems [2.174919458782602]
  In inverse problems, the incorporation of a sparsity prior yields a regularization effect on the solution.<n>We propose a probabilistic sparsity prior formulated as a mixture of Gaussians, capable of modeling sparsity with respect to a generic basis.<n>We put forth both a supervised and an unsupervised training strategy to estimate the parameters of this network.
  arXiv  Detail & Related papers  (2024-01-29T22:52:57Z)
• Regularized Linear Regression for Binary Classification [20.710343135282116]
  Regularized linear regression is a promising approach for binary classification problems in which the training set has noisy labels. We show that for large enough regularization strength, the optimal weights concentrate around two values of opposite sign. We observe that in many cases the corresponding "compression" of each weight to a single bit leads to very little loss in performance.
  arXiv  Detail & Related papers  (2023-11-03T23:18:21Z)
• The Implicit Bias of Batch Normalization in Linear Models and Two-layer Linear Convolutional Neural Networks [117.93273337740442]
  We show that gradient descent converges to a uniform margin classifier on the training data with an $exp(-Omega(log2 t))$ convergence rate. We also show that batch normalization has an implicit bias towards a patch-wise uniform margin.
  arXiv  Detail & Related papers  (2023-06-20T16:58:00Z)
• Instance-Dependent Generalization Bounds via Optimal Transport [51.71650746285469]
  Existing generalization bounds fail to explain crucial factors that drive the generalization of modern neural networks. We derive instance-dependent generalization bounds that depend on the local Lipschitz regularity of the learned prediction function in the data space. We empirically analyze our generalization bounds for neural networks, showing that the bound values are meaningful and capture the effect of popular regularization methods during training.
  arXiv  Detail & Related papers  (2022-11-02T16:39:42Z)
• Self-supervised debiasing using low rank regularization [59.84695042540525]
  Spurious correlations can cause strong biases in deep neural networks, impairing generalization ability. We propose a self-supervised debiasing framework potentially compatible with unlabeled samples. Remarkably, the proposed debiasing framework significantly improves the generalization performance of self-supervised learning baselines.
  arXiv  Detail & Related papers  (2022-10-11T08:26:19Z)
• Benign Overfitting in Adversarially Robust Linear Classification [91.42259226639837]
  "Benign overfitting", where classifiers memorize noisy training data yet still achieve a good generalization performance, has drawn great attention in the machine learning community. We show that benign overfitting indeed occurs in adversarial training, a principled approach to defend against adversarial examples.
  arXiv  Detail & Related papers  (2021-12-31T00:27:31Z)
• Information-Theoretic Generalization Bounds for Iterative Semi-Supervised Learning [81.1071978288003]
  In particular, we seek to understand the behaviour of the em generalization error of iterative SSL algorithms using information-theoretic principles. Our theoretical results suggest that when the class conditional variances are not too large, the upper bound on the generalization error decreases monotonically with the number of iterations, but quickly saturates.
  arXiv  Detail & Related papers  (2021-10-03T05:38:49Z)
• Detecting Label Noise via Leave-One-Out Cross Validation [0.0]
  We present a simple algorithm for identifying and correcting real-valued noisy labels from a mixture of clean and corrupted samples. A heteroscedastic noise model is employed, in which additive Gaussian noise terms with independent variances are associated with each and all of the observed labels. We show that the presented method can pinpoint corrupted samples and lead to better regression models when trained on synthetic and real-world scientific data sets.
  arXiv  Detail & Related papers  (2021-03-21T10:02:50Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.